I have a question regarding to the notion of using Poisson dist as an approximation to Binomial dist.
I can easily prove $\lim_{n\rightarrow\infty} P(X=k) = P(Y=k)$, where $X \sim$ Binomial(n, p) and $Y\sim$ Poisson($\lambda$).
However, I just couldn't quite get the idea of $\lambda = np$ when $n$ is sufficiently large and $p$ is sufficiently small.
Isn't $\lambda$ is the average number of "successes" over an interval?
How is $np$ the average number of "successes" over an interval?
(Note: $n$ represents the number of identical and independent trials; $p$ is the probability of "success" in one trial.)

Many thanks,

  • $\begingroup$ When you proved your limit assertion, what was $\lambda$? If $p$ is kept fixed, both limits are $0$, which tells us nothing about how the distributions compare. $\endgroup$ Jun 30 '16 at 20:22
  • $\begingroup$ @AndréNicolas $\lambda$ is the average number of "success" over an interval and what I learnt from the textbook is that we assume $\lambda = np$ , then that limit holds when $n$ are large enough and $p$ are small enough. I believe $\lambda$ is fixed, not $p$. $\endgroup$ Jun 30 '16 at 20:26
  • $\begingroup$ In a certain time interval, we perform an experiment independently $n$ times, where the probability of success is $\lambda/n$. The number of successes has binomial distribution, parameters $\lambda/n$ and $n$, so with mean $n(\lambda/n)=\lambda$. $\endgroup$ Jun 30 '16 at 20:30
  • $\begingroup$ @AndréNicolas Oh, I see. Thank you so much. $\endgroup$ Jun 30 '16 at 20:33
  • $\begingroup$ @AndréNicolas Thanks for the reply but I do know how to prove it $\endgroup$ Jul 1 '16 at 1:29

Here is the setup. In a certain interval, say of time, we perform an experiment independently $n$ times, with probability of success each time equal to $\frac{\lambda}{n}$.

Then the number of successes $X_n$ in that time interval has binomial distribution, with parameters $p_n=\frac{\lambda}{n}$ and $n$. Hence $X_n$ has mean $n\cdot\frac{\lambda}{n}$, that is, $\lambda$.

Your limit calculation presumably showed that $\lim_{n\to\infty}\Pr(X_n=k)=e^{-\lambda}\frac{\lambda^k}{k!}$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.